lockstat.c

Sun Solaris 10 中的 DTrace 组件的源代码。请参看: http://www.sun.com/software/solaris/observability.jsp

/*
 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License, Version 1.0 only.
 * See the file usr/src/LICENSING.NOTICE in this distribution or
 * http://www.opensolaris.org/license/ for details.
 */

#pragma ident	"@(#)lockstat.c	1.12	04/11/17 SMI"

#include <stdio.h>
#include <stddef.h>
#include <stdlib.h>
#include <stdarg.h>
#include <string.h>
#include <strings.h>
#include <ctype.h>
#include <fcntl.h>
#include <unistd.h>
#include <errno.h>
#include <limits.h>
#include <sys/types.h>
#include <sys/modctl.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <dtrace.h>
#include <sys/lockstat.h>
#include <alloca.h>
#include <signal.h>

#define	LS_MAX_STACK_DEPTH	50
#define	LS_MAX_EVENTS		64

typedef struct lsrec {
	struct lsrec	*ls_next;	/* next in hash chain */
	uintptr_t	ls_lock;	/* lock address */
	uintptr_t	ls_caller;	/* caller address */
	uint32_t	ls_count;	/* cumulative event count */
	uint32_t	ls_event;	/* type of event */
	uintptr_t	ls_refcnt;	/* cumulative reference count */
	uint64_t	ls_time;	/* cumulative event duration */
	uint32_t	ls_hist[64];	/* log2(duration) histogram */
	uintptr_t	ls_stack[LS_MAX_STACK_DEPTH];
} lsrec_t;

typedef struct lsdata {
	struct lsrec	*lsd_next;	/* next available */
	int		lsd_count;	/* number of records */
} lsdata_t;

/*
 * Definitions for the types of experiments which can be run.  They are
 * listed in increasing order of memory cost and processing time cost.
 * The numerical value of each type is the number of bytes needed per record.
 */
#define	LS_BASIC	offsetof(lsrec_t, ls_time)
#define	LS_TIME		offsetof(lsrec_t, ls_hist[0])
#define	LS_HIST		offsetof(lsrec_t, ls_stack[0])
#define	LS_STACK(depth)	offsetof(lsrec_t, ls_stack[depth])

static void report_stats(FILE *, lsrec_t **, size_t, uint64_t, uint64_t);
static void report_trace(FILE *, lsrec_t **);

extern int symtab_init(void);
extern char *addr_to_sym(uintptr_t, uintptr_t *, size_t *);
extern uintptr_t sym_to_addr(char *name);
extern size_t sym_size(char *name);
extern char *strtok_r(char *, const char *, char **);

#define	DEFAULT_NRECS	10000
#define	DEFAULT_HZ	97
#define	MAX_HZ		1000

static int g_stkdepth;
static int g_topn = INT_MAX;
static hrtime_t g_elapsed;
static int g_rates = 0;
static int g_pflag = 0;
static int g_Pflag = 0;
static int g_wflag = 0;
static int g_Wflag = 0;
static int g_cflag = 0;
static int g_kflag = 0;
static int g_gflag = 0;
static int g_Vflag = 0;
static int g_tracing = 0;
static size_t g_recsize;
static size_t g_nrecs;
static int g_nrecs_used;
static uchar_t g_enabled[LS_MAX_EVENTS];
static hrtime_t g_min_duration[LS_MAX_EVENTS];
static dtrace_hdl_t *g_dtp;
static char *g_predicate;
static char *g_ipredicate;
static char *g_prog;
static int g_proglen;
static int g_dropped;

typedef struct ls_event_info {
	char	ev_type;
	char	ev_lhdr[20];
	char	ev_desc[80];
	char	ev_units[10];
	char	ev_name[DTRACE_NAMELEN];
	char	*ev_predicate;
	char	*ev_acquire;
} ls_event_info_t;

static ls_event_info_t g_event_info[LS_MAX_EVENTS] = {
	{ 'C',	"Lock",	"Adaptive mutex spin",			"spin",
	    "lockstat:::adaptive-spin" },
	{ 'C',	"Lock",	"Adaptive mutex block",			"nsec",
	    "lockstat:::adaptive-block" },
	{ 'C',	"Lock",	"Spin lock spin",			"spin",
	    "lockstat:::spin-spin" },
	{ 'C',	"Lock",	"Thread lock spin",			"spin",
	    "lockstat:::thread-spin" },
	{ 'C',	"Lock",	"R/W writer blocked by writer",		"nsec",
	    "lockstat:::rw-block", "arg2 == 0 && arg3 == 1" },
	{ 'C',	"Lock",	"R/W writer blocked by readers",	"nsec",
	    "lockstat:::rw-block", "arg2 == 0 && arg3 == 0 && arg4" },
	{ 'C',	"Lock",	"R/W reader blocked by writer",		"nsec",
	    "lockstat:::rw-block", "arg2 != 0 && arg3 == 1" },
	{ 'C',	"Lock",	"R/W reader blocked by write wanted",	"nsec",
	    "lockstat:::rw-block", "arg2 != 0 && arg3 == 0 && arg4" },
	{ 'C',	"Lock",	"Unknown event (type 8)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 9)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 10)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 11)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 12)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 13)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 14)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 15)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 16)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 17)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 18)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 19)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 20)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 21)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 22)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 23)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 24)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 25)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 26)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 27)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 28)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 29)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 30)",		"units"	},
	{ 'C',	"Lock",	"Unknown event (type 31)",		"units"	},
	{ 'H',	"Lock",	"Adaptive mutex hold",			"nsec",
	    "lockstat:::adaptive-release", NULL,
	    "lockstat:::adaptive-acquire" },
	{ 'H',	"Lock",	"Spin lock hold",			"nsec",
	    "lockstat:::spin-release", NULL,
	    "lockstat:::spin-acquire" },
	{ 'H',	"Lock",	"R/W writer hold",			"nsec",
	    "lockstat:::rw-release", "arg1 == 0",
	    "lockstat:::rw-acquire" },
	{ 'H',	"Lock",	"R/W reader hold",			"nsec",
	    "lockstat:::rw-release", "arg1 != 0",
	    "lockstat:::rw-acquire" },
	{ 'H',	"Lock",	"Unknown event (type 36)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 37)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 38)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 39)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 40)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 41)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 42)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 43)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 44)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 45)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 46)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 47)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 48)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 49)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 50)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 51)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 52)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 53)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 54)",		"units"	},
	{ 'H',	"Lock",	"Unknown event (type 55)",		"units"	},
	{ 'I',	"CPU+PIL", "Profiling interrupt",		"nsec",
	    "profile:::profile-97", NULL },
	{ 'I',	"Lock",	"Unknown event (type 57)",		"units"	},
	{ 'I',	"Lock",	"Unknown event (type 58)",		"units"	},
	{ 'I',	"Lock",	"Unknown event (type 59)",		"units"	},
	{ 'E',	"Lock",	"Recursive lock entry detected",	"(N/A)",
	    "lockstat:::rw-release", NULL, "lockstat:::rw-acquire" },
	{ 'E',	"Lock",	"Lockstat enter failure",		"(N/A)"	},
	{ 'E',	"Lock",	"Lockstat exit failure",		"nsec"	},
	{ 'E',	"Lock",	"Lockstat record failure",		"(N/A)"	},
};

static void
fail(int do_perror, const char *message, ...)
{
	va_list args;
	int save_errno = errno;

	va_start(args, message);
	(void) fprintf(stderr, "lockstat: ");
	(void) vfprintf(stderr, message, args);
	va_end(args);
	if (do_perror)
		(void) fprintf(stderr, ": %s", strerror(save_errno));
	(void) fprintf(stderr, "\n");
	exit(2);
}

static void
dfail(const char *message, ...)
{
	va_list args;

	va_start(args, message);
	(void) fprintf(stderr, "lockstat: ");
	(void) vfprintf(stderr, message, args);
	va_end(args);
	(void) fprintf(stderr, ": %s\n",
	    dtrace_errmsg(g_dtp, dtrace_errno(g_dtp)));

	exit(2);
}

static void
show_events(char event_type, char *desc)
{
	int i, first = -1, last;

	for (i = 0; i < LS_MAX_EVENTS; i++) {
		ls_event_info_t *evp = &g_event_info[i];
		if (evp->ev_type != event_type ||
		    strncmp(evp->ev_desc, "Unknown event", 13) == 0)
			continue;
		if (first == -1)
			first = i;
		last = i;
	}

	(void) fprintf(stderr,
	    "\n%s events (lockstat -%c or lockstat -e %d-%d):\n\n",
	    desc, event_type, first, last);

	for (i = first; i <= last; i++)
		(void) fprintf(stderr,
		    "%4d = %s\n", i, g_event_info[i].ev_desc);
}

static void
usage(void)
{
	(void) fprintf(stderr,
	    "Usage: lockstat [options] command [args]\n"
	    "\nEvent selection options:\n\n"
	    "  -C              watch contention events [on by default]\n"
	    "  -E              watch error events [off by default]\n"
	    "  -H              watch hold events [off by default]\n"
	    "  -I              watch interrupt events [off by default]\n"
	    "  -A              watch all lock events [equivalent to -CH]\n"
	    "  -e event_list   only watch the specified events (shown below);\n"
	    "                  <event_list> is a comma-separated list of\n"
	    "                  events or ranges of events, e.g. 1,4-7,35\n"
	    "  -i rate         interrupt rate for -I [default: %d Hz]\n"
	    "\nData gathering options:\n\n"
	    "  -b              basic statistics (lock, caller, event count)\n"
	    "  -t              timing for all events [default]\n"
	    "  -h              histograms for event times\n"
	    "  -s depth        stack traces <depth> deep\n"
	    "\nData filtering options:\n\n"
	    "  -n nrecords     maximum number of data records [default: %d]\n"
	    "  -l lock[,size]  only watch <lock>, which can be specified as a\n"
	    "                  symbolic name or hex address; <size> defaults\n"
	    "                  to the ELF symbol size if available, 1 if not\n"
	    "  -f func[,size]  only watch events generated by <func>\n"
	    "  -d duration     only watch events longer than <duration>\n"
	    "  -T              trace (rather than sample) events\n"
	    "\nData reporting options:\n\n"
	    "  -c              coalesce lock data for arrays like pse_mutex[]\n"
	    "  -k              coalesce PCs within functions\n"
	    "  -g              show total events generated by function\n"
	    "  -w              wherever: don't distinguish events by caller\n"
	    "  -W              whichever: don't distinguish events by lock\n"
	    "  -R              display rates rather than counts\n"
	    "  -p              parsable output format (awk(1)-friendly)\n"
	    "  -P              sort lock data by (count * avg_time) product\n"
	    "  -D n            only display top <n> events of each type\n"
	    "  -o filename     send output to <filename>\n",
	    DEFAULT_HZ, DEFAULT_NRECS);

	show_events('C', "Contention");
	show_events('H', "Hold-time");
	show_events('I', "Interrupt");
	show_events('E', "Error");
	(void) fprintf(stderr, "\n");

	exit(1);
}

static int
lockcmp(lsrec_t *a, lsrec_t *b)
{
	int i;

	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	for (i = g_stkdepth - 1; i >= 0; i--) {
		if (a->ls_stack[i] < b->ls_stack[i])
			return (-1);
		if (a->ls_stack[i] > b->ls_stack[i])
			return (1);
	}

	if (a->ls_caller < b->ls_caller)
		return (-1);
	if (a->ls_caller > b->ls_caller)
		return (1);

	if (a->ls_lock < b->ls_lock)
		return (-1);
	if (a->ls_lock > b->ls_lock)
		return (1);

	return (0);
}

static int
countcmp(lsrec_t *a, lsrec_t *b)
{
	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	return (b->ls_count - a->ls_count);
}

static int
timecmp(lsrec_t *a, lsrec_t *b)
{
	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	if (a->ls_time < b->ls_time)
		return (1);
	if (a->ls_time > b->ls_time)
		return (-1);

	return (0);
}

static int
lockcmp_anywhere(lsrec_t *a, lsrec_t *b)
{
	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	if (a->ls_lock < b->ls_lock)
		return (-1);
	if (a->ls_lock > b->ls_lock)
		return (1);

	return (0);
}

static int
lock_and_count_cmp_anywhere(lsrec_t *a, lsrec_t *b)
{
	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	if (a->ls_lock < b->ls_lock)
		return (-1);
	if (a->ls_lock > b->ls_lock)
		return (1);

	return (b->ls_count - a->ls_count);
}

static int
sitecmp_anylock(lsrec_t *a, lsrec_t *b)
{
	int i;

	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	for (i = g_stkdepth - 1; i >= 0; i--) {
		if (a->ls_stack[i] < b->ls_stack[i])
			return (-1);
		if (a->ls_stack[i] > b->ls_stack[i])
			return (1);
	}

	if (a->ls_caller < b->ls_caller)
		return (-1);
	if (a->ls_caller > b->ls_caller)
		return (1);

	return (0);
}

static int
site_and_count_cmp_anylock(lsrec_t *a, lsrec_t *b)
{
	int i;

	if (a->ls_event < b->ls_event)
		return (-1);
	if (a->ls_event > b->ls_event)
		return (1);

	for (i = g_stkdepth - 1; i >= 0; i--) {
		if (a->ls_stack[i] < b->ls_stack[i])
			return (-1);
		if (a->ls_stack[i] > b->ls_stack[i])
			return (1);
	}

	if (a->ls_caller < b->ls_caller)
		return (-1);
	if (a->ls_caller > b->ls_caller)
		return (1);

	return (b->ls_count - a->ls_count);
}

static void
mergesort(int (*cmp)(lsrec_t *, lsrec_t *), lsrec_t **a, lsrec_t **b, int n)
{
	int m = n / 2;
	int i, j;

	if (m > 1)
		mergesort(cmp, a, b, m);
	if (n - m > 1)
		mergesort(cmp, a + m, b + m, n - m);
	for (i = m; i > 0; i--)
		b[i - 1] = a[i - 1];
	for (j = m - 1; j < n - 1; j++)
		b[n + m - j - 2] = a[j + 1];
	while (i < j)
		*a++ = cmp(b[i], b[j]) < 0 ? b[i++] : b[j--];
	*a = b[i];
}

static void
coalesce(int (*cmp)(lsrec_t *, lsrec_t *), lsrec_t **lock, int n)
{
	int i, j;
	lsrec_t *target, *current;

	target = lock[0];

	for (i = 1; i < n; i++) {
		current = lock[i];
		if (cmp(current, target) != 0) {
			target = current;
			continue;
		}
		current->ls_event = LS_MAX_EVENTS;
		target->ls_count += current->ls_count;
		target->ls_refcnt += current->ls_refcnt;
		if (g_recsize < LS_TIME)
			continue;
		target->ls_time += current->ls_time;
		if (g_recsize < LS_HIST)
			continue;
		for (j = 0; j < 64; j++)
			target->ls_hist[j] += current->ls_hist[j];
	}
}

static void
coalesce_symbol(uintptr_t *addrp)
{
	uintptr_t symoff;
	size_t symsize;

	if (addr_to_sym(*addrp, &symoff, &symsize) != NULL && symoff < symsize)
		*addrp -= symoff;
}

static void
predicate_add(char **pred, char *what, char *cmp, uintptr_t value)
{
	char *new;
	int len, newlen;

	if (what == NULL)
		return;

	if (*pred == NULL) {
		*pred = malloc(1);
		*pred[0] = '\0';
	}

	len = strlen(*pred);
	newlen = len + strlen(what) + 32 + strlen("( && )");
	new = malloc(newlen);

	if (*pred[0] != '\0') {
		if (cmp != NULL) {
			(void) sprintf(new, "(%s) && (%s %s 0x%p)",
			    *pred, what, cmp, (void *)value);
		} else {
			(void) sprintf(new, "(%s) && (%s)", *pred, what);
		}
	} else {
		if (cmp != NULL) {
			(void) sprintf(new, "%s %s 0x%p",
			    what, cmp, (void *)value);
		} else {
			(void) sprintf(new, "%s", what);
		}
	}

	free(*pred);
	*pred = new;
}

static void
predicate_destroy(char **pred)
{
	free(*pred);
	*pred = NULL;
}

static void
filter_add(char **filt, char *what, uintptr_t base, uintptr_t size)
{
	char buf[256], *c = buf, *new;
	int len, newlen;

	if (*filt == NULL) {
		*filt = malloc(1);
		*filt[0] = '\0';
	}

	(void) sprintf(c, "%s(%s >= 0x%p && %s < 0x%p)", *filt[0] != '\0' ?
	    " || " : "", what, (void *)base, what, (void *)(base + size));

	newlen = (len = strlen(*filt) + 1) + strlen(c);
	new = malloc(newlen);
	bcopy(*filt, new, len);
	(void) strcat(new, c);
	free(*filt);
	*filt = new;
}

static void
filter_destroy(char **filt)
{
	free(*filt);
	*filt = NULL;
}

static void
dprog_add(const char *fmt, ...)
{
	va_list args;
	int size, offs;
	char c;

	va_start(args, fmt);
	size = vsnprintf(&c, 1, fmt, args) + 1;

	if (g_proglen == 0) {